Empirical evaluation of the influence of control selection schemes on relative risk estimation: The Welsh nickel workers study

Abstract

Objective - To perform an empirical evaluation of the theory that relative incidence rate (RIR) and relative risk (RR) can be directly estimated from case-control studies that have different sampling schemes of controls. Methods - With data from the South Wales nickel refinery workers (SWNRW) study, a nested case-control study of the relation of nickel exposure to respiratory cancers, was conducted within each of four fixed subcohorts that differed for stability of exposure, incidence rates and RIR. Odds ratios (ORs) and confidence intervals (CIs) were estimated either with all available controls or with randomly sampled subsets of controls. Results - Respiratory cancers were not rare as risk of nasal and lung cancer in workers unexposed to nickel varied from 15% to 26% over the full risk period. The RIR was adequately estimated by the OR when controls were identified concurrently to case occurrence throughout the risk period. The RR was well approximated with the OR when controls were a sample of the study base. Conclusions - These results add empirical support to the theory that the RIR or the RR can be validly estimated in case-control studies. Overall, this theory is relatively tolerant of large departures from the stability assumptions of exposure and of incidence.